From 815af82b74fa0901e818f5d16ee418675f399101 Mon Sep 17 00:00:00 2001 From: Nadav Rotem Date: Wed, 19 Oct 2011 20:43:16 +0000 Subject: Improve code generation for vselect on SSE2: When checking the availability of instructions using the TLI, a 'promoted' instruction IS available. It means that the value is bitcasted to another type for which there is an operation. The correct check for the availablity of an instruction is to check if it should be expanded. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@142542 91177308-0d34-0410-b5e6-96231b3b80d8 --- test/CodeGen/X86/sse2-blend.ll | 17 +++++++++++------ 1 file changed, 11 insertions(+), 6 deletions(-) (limited to 'test/CodeGen/X86/sse2-blend.ll') diff --git a/test/CodeGen/X86/sse2-blend.ll b/test/CodeGen/X86/sse2-blend.ll index 56b099e..0007cab 100644 --- a/test/CodeGen/X86/sse2-blend.ll +++ b/test/CodeGen/X86/sse2-blend.ll @@ -1,8 +1,10 @@ -; RUN: llc < %s -march=x86 -mcpu=yonah -promote-elements -mattr=+sse2,-sse41 | FileCheck %s - - -; currently (xor v4i32) is defined as illegal, so we scalarize the code. +; RUN: llc < %s -march=x86 -mcpu=yonah -mattr=+sse2,-sse41 | FileCheck %s +; CHECK: vsel_float +; CHECK: pandn +; CHECK: pand +; CHECK: por +; CHECK: ret define void@vsel_float(<4 x float>* %v1, <4 x float>* %v2) { %A = load <4 x float>* %v1 %B = load <4 x float>* %v2 @@ -11,8 +13,11 @@ define void@vsel_float(<4 x float>* %v1, <4 x float>* %v2) { ret void } -; currently (xor v4i32) is defined as illegal, so we scalarize the code. - +; CHECK: vsel_i32 +; CHECK: pandn +; CHECK: pand +; CHECK: por +; CHECK: ret define void@vsel_i32(<4 x i32>* %v1, <4 x i32>* %v2) { %A = load <4 x i32>* %v1 %B = load <4 x i32>* %v2 -- cgit v1.1